GPT-6.1 Sol 在 Codex 与 API 的速度和额度实测

Youssof Al Toukhi @Youssofal_原推发布于 2026.10.09

在 X 上打开原推

I tested GPT 6.1 Sol in Codex vs API with Ultrafast. Ultrafast is actually marginally faster in Codex vs the API. Within variance. However, on regular fast mode 6.1 Sol is half as fast in codex vs API (42 vs 88 TPS) The BIGGEST problem is the usage. OpenAI claimed 20% higher usage than regular Astra. In truth it is 60%. In Codex, ultrafast uses limits 8x faster rather than the 6x API cost. This means 6.1 Sol ultrafast = 1.6x usage of Astra regular.